CVC Workshop Notes

Introduction to R and R studio

Scatterplot with birth data:

library(mosaic)
library(mosaicData)
xyplot(births ~ dayofyear, data = Births78)

Other commands we’ve run during the workshop day 1 and some comments on them.

histogram(~ age, data = HELPrct) #histogram of age

favstats(~ age, data = HELPrct) #My favorite statistics; it ignores NAs

##  min Q1 median Q3 max     mean       sd   n missing
##   19 30     35 40  60 35.65342 7.710266 453       0

tally(~sex, data = HELPrct) #Count of gender

## 
## female   male 
##    107    346

tally(~sex, format = "percent", data = HELPrct) #Percents of gender

## 
##   female     male 
## 23.62031 76.37969

tally(~sex, format = "proportion", data = HELPrct) #Proportions of gender

## 
##    female      male 
## 0.2362031 0.7637969

tally(~substance, data = HELPrct) #Count of substance

## 
## alcohol cocaine  heroin 
##     177     152     124

tally(~substance, format = "perc", data = HELPrct) #Percents of substance

## 
##  alcohol  cocaine   heroin 
## 39.07285 33.55408 27.37307

tally(sex ~ substance, data = HELPrct) #Cross-tab sex & substance

##         substance
## sex      alcohol cocaine heroin
##   female      36      41     30
##   male       141     111     94

tally(~ sex + substance, data = HELPrct) #Ditto, just a different format

##         substance
## sex      alcohol cocaine heroin
##   female      36      41     30
##   male       141     111     94

Also, always remember the mplot command for producing graphs and clicking on the ‘Show Expression’.

Data Structures and Tidy Data

Consider cases and variables

a row is always a case
a column is always a variable

In tidy data:

No units
No footnotes

All of this goes into a codebook only.

Visualizing Data

DTK: * Glyphs are marks; data glyps are also marks and the features of the glyphs encode the values of the variables & the visual properties are aesthetics. The choices that we make as experts in our field is the choice of aesthetic to map to variables. The word ‘aesthetic’ here is taken from its early When we make data glyps we map variables to aesthetics. A scale is for a computer and a guide is for a person (about the aesthetics). A legend (beside a graph) is an example of a guide. A data table is glyph-ready when there is one row for each glyp to be drawn (I could say, “that’s x-position, that’s y-position, that’s color, that’s size.” Glyph-ready data are tidy data, but tidy data are not necessarily glyph-ready. *Sometimes glyphs represent the collective properties of variables, e.g. in the case of histograms.

RP:

Teaches going from the histogram back to the data, rather than going from the data to the histogram because of having had the experience of studnets rote learning to go from data to a histogram without actually understanding what histograms mean.

require(lubridate)
data(Births78)
head(Births78, 3)

##         date births dayofyear
## 1 1978-01-01   7701         1
## 2 1978-01-02   7527         2
## 3 1978-01-03   8825         3

We want to generate a plot using ggplot() + geom_point()
We also need aesthetics, aes(), e.g. aes(x = date, y = births)
And a call on the data, data = Births78

ggplot(data = Births78, aes(x = date, y = births)) + geom_point()

But, we need to add days of the week because that’s more useful to us:

Births78 <- 
  Births78 %>% 
  mutate(wday = wday(date, label = TRUE))
  ggplot(data = Births78, aes(x = date, y = births, color = wday)) + geom_point()

Note, that the same graph would be generated by the following:

  ggplot(data = Births78) + geom_point(aes(x = date, y = births, color = wday))

We could change this to a line graph:

 ggplot(data = Births78) + geom_line(aes(x = date, y = births, color = wday))

Or we could have points and lines; note that we have moved the aes commands into the ggplot command because it applies to all the layers.

 ggplot(data = Births78, aes(x = date, y = births, color = wday)) + geom_line() + geom_point()

We could have put the data outside the ggplot command using the magrittr pipe:

 Births78 %>%
  ggplot(aes(x = date, y = births, color = wday)) + geom_line() + geom_point()

We need to do setting rather than mapping for things like colors; inside the individual geom you set the color you want (different to ggvis where setting is done with :=). Within ggplot you can only map nor set.

 Births78 %>%
  ggplot(aes(x = date, y = births)) + geom_point(color = "navy")

Combine the colored lines with navy points - notice that wday is in the aes in the geom_line whereas with geom_point we don’t call on aes because we’re simply setting (not mapping).

 Births78 %>%
  ggplot(aes(x = date, y = births)) + 
    geom_line(aes(color = wday)) + 
    geom_point(color = "navy")

Recall that we can check out the kinds of geoms that exist if we use the command apropos, notice the use of the caret (^)

apropos("^geom")

##  [1] "geom_abline"     "geom_area"       "geom_bar"       
##  [4] "geom_bin2d"      "geom_blank"      "geom_boxplot"   
##  [7] "geom_contour"    "geom_crossbar"   "geom_density"   
## [10] "geom_density2d"  "geom_dotplot"    "geom_errorbar"  
## [13] "geom_errorbarh"  "geom_freqpoly"   "geom_hex"       
## [16] "geom_histogram"  "geom_hline"      "geom_jitter"    
## [19] "geom_line"       "geom_linerange"  "geom_map"       
## [22] "geom_path"       "geom_point"      "geom_pointrange"
## [25] "geom_polygon"    "geom_quantile"   "geom_raster"    
## [28] "geom_rect"       "geom_ribbon"     "geom_rug"       
## [31] "geom_segment"    "geom_smooth"     "geom_step"      
## [34] "geom_text"       "geom_tile"       "geom_violin"    
## [37] "geom_vline"

Key plot metric: Does my plot make the comparisions I am interested in easily and accurately.
We can generate a new geom, a geom_bar:

HELPrct %>% 
ggplot(aes(x = substance)) + 
geom_bar()

Notice that we were able to construct the bar chart even though our data weren’t glyph-ready with counts; ggplot did it for us.

We want to generate histograms where we can set the binwidth:

HELPrct %>% 
    ggplot(aes(x = age)) + 
    geom_histogram(binwidth = 2)

* We also often want to use frequency polygons or kernel density functions:

HELPrct %>% 
ggplot(aes(x = age)) + 
    geom_freqpoly(binwidth = 2)

HELPrct %>% 
ggplot(aes(x = age)) + 
    geom_density()

But, note RP prefers to add density to a line plot because it look

HELPrct %>% 
ggplot(aes(x = age)) + 
geom_line(stat = "density")

Or we could have put the geom in the stat_density

HELPrct %>%
ggplot(aes(x=age)) +
stat_density( geom="line")

## ymax not defined: adjusting position using y instead

Now generate your own graph looking at the average consumption of graphs (which I did by groups)

HELPrct %>% 
    ggplot(aes(x = i1)) + 
    geom_line(stat = "density", aes(color = factor(substance)))

Data Wrangling

We want to break the data wrangling up into small steps with simple rules to deal with simple ‘chunks’ at each point.
We shall start with a sub-set of commands that are most useful and ignore much of the many things we often use when learning R (if you have familiarity), e.g. we’ll ignore $, [], [[]],

There are 5 kinds of things (objects) in R:

data tables (aka dataframes), our convention is for first letter upper-case, e.g. BabyNames
variables, i.e. the stuff in the tables,our convention is for first letter lower-case, e.g. year
scalars, e.g. "Treatment" or 42 (thanks DA)
functions & their arguments: e.g. positional arguments, named arguments (parentheses follow names of things if and only if those things are functions, e.g. sd(); multiple arguments separated with commas)
pipes: arrange sequences, %>% takes something and puts it as the/an input to a function, e.g. something >%> function() rather than function(something, other stuff, ...).

3 Main kinds of functions for Wrangling

Data verb takes a data table as input, does something & gives a data table as output
Transformation function takes a variable and transforms it in some way; takes a variable as input and produces a variable as output (a la Sraffa?)
Reduction function takes a bunch of variables and turns them into a single variable, e.g. the mean, max, standard deviation, etc

To run the following you may need to install the packages NHANES, dplyr, and babynames. Run the command, e.g. install.packages("NHANES"). Remember to require these packages.

Now we run some commands to start wrangling the data.

require(NHANES)
require(dplyr)
require(babynames)
NHANESmajors <- filter(NHANES, Age >= 21)
nowsmoking <- select(NHANES, Age, SmokeNow)
nowsmoking %>% 
  head()

## Source: local data frame [6 x 2]
## 
##   Age SmokeNow
## 1  34       No
## 2  34       No
## 3  34       No
## 4   4       NA
## 5  49      Yes
## 6   9       NA

group_by(babynames, name, year)

## Source: local data frame [1,792,091 x 5]
## Groups: name, year
## 
##    year sex      name    n       prop
## 1  1880   F      Mary 7065 0.07238359
## 2  1880   F      Anna 2604 0.02667896
## 3  1880   F      Emma 2003 0.02052149
## 4  1880   F Elizabeth 1939 0.01986579
## 5  1880   F    Minnie 1746 0.01788843
## 6  1880   F  Margaret 1578 0.01616720
## 7  1880   F       Ida 1472 0.01508119
## 8  1880   F     Alice 1414 0.01448696
## 9  1880   F    Bertha 1320 0.01352390
## 10 1880   F     Sarah 1288 0.01319605
## ..  ... ...       ...  ...        ...

head(KidsFeet)

##     name birthmonth birthyear length width sex biggerfoot domhand
## 1  David          5        88   24.4   8.4   B          L       R
## 2   Lars         10        87   25.4   8.8   B          L       L
## 3   Zach         12        87   24.5   9.7   B          R       R
## 4   Josh          1        88   25.2   9.8   B          L       R
## 5   Lang          2        88   25.1   8.9   B          L       R
## 6 Scotty          3        88   25.7   9.7   B          R       R

select(KidsFeet, birthyear, domhand) %>% 
  head()

##   birthyear domhand
## 1        88       R
## 2        87       L
## 3        87       R
## 4        88       R
## 5        88       R
## 6        88       R

righties <- filter(KidsFeet, domhand == "R")
head(righties)

##     name birthmonth birthyear length width sex biggerfoot domhand
## 1  David          5        88   24.4   8.4   B          L       R
## 2   Zach         12        87   24.5   9.7   B          R       R
## 3   Josh          1        88   25.2   9.8   B          L       R
## 4   Lang          2        88   25.1   8.9   B          L       R
## 5 Scotty          3        88   25.7   9.7   B          R       R
## 6 Edward          2        88   26.1   9.6   B          L       R

The pipe (%>%) makes it clear what the individual lines are doing.

KidsFeetArea <- KidsFeet %>% 
  mutate(area = length * width) %>% 
  head()

arrange orders your cases in terms of the variable you select in ascending order, e.g. arrange(area).

KidsFeet %>% 
  mutate(area = length * width) %>% 
  arrange(area) %>% 
  head()

##      name birthmonth birthyear length width sex biggerfoot domhand   area
## 1  Hayley          1        88   21.6   7.9   G          R       R 170.64
## 2    Kate          4        88   23.7   7.9   G          R       R 187.23
## 3 Caitlin          7        88   22.5   8.6   G          R       R 193.50
## 4  Hannah          3        88   22.9   8.5   G          L       R 194.65
## 5   Peggy         10        88   24.2   8.1   G          L       R 196.02
## 6   Laura          9        88   24.0   8.3   G          R       L 199.20

To arrange in descending order, you can add desc() within arrange.

KidsFeet %>% 
  mutate(area = length * width) %>% 
  arrange(desc(area)) %>% 
  head()

##     name birthmonth birthyear length width sex biggerfoot domhand   area
## 1   Mark          9        87   27.5   9.8   B          R       R 269.50
## 2    Cam          3        88   27.0   9.8   B          L       R 264.60
## 3   Glen          7        88   27.1   9.4   B          L       R 254.74
## 4 Edward          2        88   26.1   9.6   B          L       R 250.56
## 5 Scotty          3        88   25.7   9.7   B          R       R 249.29
## 6   Abby          2        88   26.1   9.5   G          L       R 247.95

We can also add another argument to arrange to break ties, e.g. adding birthmonth (I’m going to head for 10 cases):

KidsFeet %>% 
  mutate(area = length * width) %>% 
  arrange(area, desc(birthmonth)) %>% 
  head(10)

##        name birthmonth birthyear length width sex biggerfoot domhand
## 1    Hayley          1        88   21.6   7.9   G          R       R
## 2      Kate          4        88   23.7   7.9   G          R       R
## 3   Caitlin          7        88   22.5   8.6   G          R       R
## 4    Hannah          3        88   22.9   8.5   G          L       R
## 5     Peggy         10        88   24.2   8.1   G          L       R
## 6     Laura          9        88   24.0   8.3   G          R       L
## 7     Damon          9        88   22.9   8.8   B          R       L
## 8   Caitlin          6        88   23.0   8.8   G          L       R
## 9     David          5        88   24.4   8.4   B          L       R
## 10 Caroline         12        87   24.0   8.7   G          R       L
##      area
## 1  170.64
## 2  187.23
## 3  193.50
## 4  194.65
## 5  196.02
## 6  199.20
## 7  201.52
## 8  202.40
## 9  204.96
## 10 208.80

The Verb summarise creates a row with summary statistics that you select by defining them in functions with particular labels.

NewThing <- KidsFeet %>% 
  mutate(area = length * width)
AnotherNewThing <- NewThing %>% 
  summarise(ave_area = mean(area))
head(NewThing)

##     name birthmonth birthyear length width sex biggerfoot domhand   area
## 1  David          5        88   24.4   8.4   B          L       R 204.96
## 2   Lars         10        87   25.4   8.8   B          L       L 223.52
## 3   Zach         12        87   24.5   9.7   B          R       R 237.65
## 4   Josh          1        88   25.2   9.8   B          L       R 246.96
## 5   Lang          2        88   25.1   8.9   B          L       R 223.39
## 6 Scotty          3        88   25.7   9.7   B          R       R 249.29

head(AnotherNewThing)

##   ave_area
## 1 222.7369

The important feature of summarise is that it’ll create one row with a variety of variables that you specify, i.e. other means, say the max of the variable, e.g.

AnotherNewThing <- NewThing %>% 
  summarise(ave_area = mean(area), max_area = max(area))
AnotherNewThing

##   ave_area max_area
## 1 222.7369    269.5

We have looked at 5 commands so far:

filter
select
mutate
arrange
summarise

We shall add another, more complex command, group_by.

Let us

NewThing %>% 
  group_by(sex) %>%
  summarise(ave_area = mean(area), max_area = max(area))

## Source: local data frame [2 x 3]
## 
##   sex ave_area max_area
## 1   B 231.0140   269.50
## 2   G 214.0242   247.95

Or we could just use KidsFeet rather than assigning things to dataframes like we did with NewThing.

KidsFeet %>% 
  mutate(area = length * width) %>% 
  group_by(sex) %>%
  summarise(ave_area = mean(area), max_area = max(area))

## Source: local data frame [2 x 3]
## 
##   sex ave_area max_area
## 1   B 231.0140   269.50
## 2   G 214.0242   247.95

Or we could arrange by both dominant hand and birth year:

KidsFeet %>% 
  mutate(area = length * width) %>% 
  group_by(domhand, birthyear) %>%
  summarise(ave_area = mean(area), max_area = max(area))

## Source: local data frame [4 x 4]
## Groups: domhand
## 
##   domhand birthyear ave_area max_area
## 1       L        87 216.1600   223.52
## 2       L        88 214.6850   240.30
## 3       R        87 245.7420   269.50
## 4       R        88 220.6769   264.60

It’s important to remain aware of the change in the cases when you use group_by. 1. You have changed the cases from the cases before group_by, e.g. the kids, and then you create a new case, e.g. a group of kids. 2. Not all the variables in the original data will appear in the output of group_by, the ones you group by and the ones you create will be there.

Teaching Tips

These are tips I picked out (thus idiosyncratic):

For teaching the idea of the command line and the script: How many of you had a TI calculator? (TI = Texas Instruments; I did not know this abbreviation – like the international students) Did you know how to use it on the very first day? What about on your final day of high school? (RP)
When R outputs the error object '[object name]' not found that normally means you (or your student) have mis-spelt the variable name or dataframe name. (RP)
Pick your favorite way (command) for doing a job, don’t show students too many methods. You’ll overwhelm them if you give them too much. Less volume, more creativity. (RP)
Call tildes ~ ‘wiggles’! Sounds funny and is less scary then tilde. (RP)
Two most important questions: What do you want R to do? (goal) What must R know to do that? (RP) [me - why don’t I call R a specific name? e.g. R = Rachel - What do I want Rachel to do? What must Rachel know to do that?]
Don’t teach them all the bells & whistles in a lecture when you want it. Rather, do it when they ask. (RP)
Get students to adopt a workflow such that students can reproduce their work again with new data. (NH)
When you start a new Markdown file (Rmd), it means that you need to load each package in the Rmd file. (NH) Often it’s useful to do so with the message include = FALSE to suppress the inclusion of that code in the output file.
Students struggle to understand the ecological paradox (DTK). (relates to the fallacy of division; which is the complement of the fallacy of composition).
To explain the magrittr pipe – which we should read it as ‘then’ or ‘and then’ (taking the advice of Hadley and Garrett G). Or, we can talk about actual plumbing and piping: put some water (or a data table) in one end and send it out another.

Commonplace Book

RP: “If your students find R to be the hardest part of your course, your R is too hard and your questions are too easy.”
RP: “You need to learn to see before you can see to learn.” (As advice to professors to get them over the problem of assumed prior learning)

Working with some of my own data

Importing the medicare data…

#medicare <- read.csv("/Users/shalliday/Google Drive/simondhalliday.github.io/cvc_workshop/medicare_fy2013.csv")
medicare <- read.csv("medicare_fy2013.csv")
#head(medicare)
#str(medicare)

Trying to embed a badly thought-out plot using the ggvis package:

library(dplyr)
library(ggvis)
medicare %>% 
  ggvis(~factor(Provider.State), ~Average.Covered.Charges) %>%
  layer_bars() %>%
   add_axis("x", properties = axis_props(
    labels = list(angle = 45, align = "left", fontSize = 10)
   ))